Monday 31 May 2010

Proceduralism: Part One (Revision)

In the last few months a milestone has been passed on this blog - or more correctly surpassed. The PCG wiki which I maintain, and every so slowly expand on, now regularly gets more traffic than Ascii Dreams - despite my active solicitation of links to articles on the blog. Of everything I've written, 'Death of the Level Designer' has continued to be the most read article I've done and attracts thousands of impressions each month despite its age. But because of that age, it has become a little outdated, and in need of a refresh - especially in the light of some of the recent developments in the field of procedural content generation.

Despite the title, I wrote 'Death of the Level Designer' as an optimistic look at what the fruits of procedural content generation could be - before Spore was released and as a way of looking at that game, but also beyond it. Unfortunately, Spore wasn't the revolutionary game that many people hoped it would be - but it did deliver in many ways a new, powerful procedural content model which no other game has since matched.

At the same time, I saw a need to try to better define what procedural content generation was: and the taxonomy I proposed was in part a reaction to what I termed little-p procedural generation such as procedural animation and procedural textures. I argued this procedural generation did not directly impact game play, while dominating the definition of procedural generation in places such as Wikipedia, to the detriment of procedural content generation of levels, plots and other game focused content. As is always the case when you position a definition in opposition to something, that definition loses a level of sophistication; and I've modified my view, and more importantly the editorial content of the PCG wiki, to more correctly reflect procedural content as a continuum of procedural generation techniques, from procedural graphics to maze generation algorithms, world building and beyond.

What I see as increased interest in procedural content generation - driven in part as a way of providing more varied content in triple A titles, but mostly by the independent development community as a way of cheaply generating enough content for their games to distinguish themselves in a crowded field - hasn't necessarily translated into more sophisticated ways of generating this content.

The same techniques of 2d height maps, tile-based maze generation and random loot drops have been used in multiple different formats (Spelunky, Torchlight, Borderlands), without expanding into true 3d or utilising procedural narrative or plot generation in ways promised by games like Facade or Balance of Power. That's not to say there hasn't been a number of exciting innovations: chief of which is the AI Director from Left4Dead 1 & 2, the free form play of Far Cry 2 and roguelike-like revisionism of Spelunky, and the contract bridge style bidding system in Frozen Synapse, but no one has released a grand procedural game in the vein of Spore (except Eskil Steenberg's Love*) or pushed the boundaries that I expected to be broken.

There have been a huge number of games which have incorporated procedural generation in interesting ways without being truly innovative in its use: the TIG Source procedural generation competition was highlight for me, the 7 day rogue-like (7DRL) competition has grown every year, Eufloria, Canabalt and other indie titles have capitalized on procedural techniques and the excitement about up-and-coming games like Subversion and Infinity: the Quest for Earth is palpable. But procedural content generation is being trumped by user content generation in making the transition from 2d to 3d (Minecraft, 3D Dot Heroes) and building new narrative techniques (Spore: Space Adventures, Sleep is Death), in ways which appear limited by its reliance on repetition and randomness.

And procedural content driven games leave themselves open to the same rehashed criticism that is brought up every time they are mentioned in the gaming forums: the content created is repetitive, predictable and easily exploited, and clearly inferior to designs created by hand, an experience directly contrary to what I've found in the rogue-like genre.

So this article series is an attempt to update and expand the discussion around procedural content generation, take a look at more sophisticated ways of classifying procedural content, see what algorithms and techniques are potential low hanging fruit to try to guess where procedural content generation in games is going, but more importantly, hazard a better guess at where these games can potentially end up. In part two, I'll start by looking at new taxonomies.

*I'd be interested in hearing your experiences of Love.


Andrew Doull said...

I haven't forgotten about the Quest for Quests - but the bite size chunks of time I have at the moment don't lend themselves well to writing the large article the next Quests part deserves.

Stromko said...

Love was really quite fantastic when I had gotten into it, designing cities with a crew of people that I grew to knew and respect, discovering things about the world every day. My time with it ended a bit abruptly when it lost my account, and with it my randomly generated name. No longer would I be Charon on Atlanta 1, I'd have to start from scratch, and I was too busy with other things to start again from nothing.

Richard said...

... the content created is repetitive, predictable and easily exploited, and clearly inferior to designs created by hand, an experience directly contrary to what I've found in the rogue-like genre.

The depth of a roguelike game is bounded by how a hodge-podge of random elements are combined. And when I say random, I do not just mean procedural generation, I also mean that the designers chose to add into the mix. Fantasy elements dragged in from here and there, modern concepts and elements dragged in elsewhere. Shops in the middle of a never ending dungeon? How did this huge place come to be? Roguelike games are no better than any other genre. Sure their gameplay may leave the railroad tracks that other asset driven genres are bound to, but that only goes so far.

This was highlighted for me recently when I played Incursion. I think Incursion is a wonderful product, but reading the manifesto on the web site it alludes to rising above the shallowness that riddles the roguelike genre. The same things I list above. But yet it still has dungeons that make no sense, random yet interesting rooms linked together via tunnels with little regard for what kind of world this makes.

I would love to see innovation in use of procedural generation, but as someone who entered the 7DRL competition, I think the barrier to entry is too high. Anyone choosing to make use of it has to start from scratch. Take for instance dungeon generation, I read no end of tutorials about the different selection of approaches people had written about. I was left knowing how to take their approaches and generate dungeons that looked like theirs, or a simple approximation they chose to write about. But to really generate dungeons I would find interesting, I might as well ignore 99% of what I have read and just experiment until I had explored the options, and pretty much started from scratch.

Unknown said...

Really looking forward to this article series and seeing which aspects of procedural generation you will discuss.